The two language problem in vectorized python expressions

About this talk

There is a bit of a misleading conception that in order to make python code fast the best approach is to handle code in vectorized expressions. While this can be true in some cases, especially when your code can be mapped to an exact function call of numpy/scipy it is a dangerous path if followed without critical thinking. Algorithms can increase the complexity when vectorizing code. This can happen without the user being aware of it, since vectorized code is using other libraries. In this talk we will iterate over examples on how to think about devectorized expressions using cython and discuss the two language problem that arises in such a setting.